diff options
Diffstat (limited to 'TAO/tests/Connection_Purging/README')
| -rw-r--r-- | TAO/tests/Connection_Purging/README | 108 |
1 files changed, 0 insertions, 108 deletions
diff --git a/TAO/tests/Connection_Purging/README b/TAO/tests/Connection_Purging/README deleted file mode 100644 index 2af5b864236..00000000000 --- a/TAO/tests/Connection_Purging/README +++ /dev/null @@ -1,108 +0,0 @@ -Connection Purging Test By: Kirthika Parameswaran <kirthika@cs.wustl.edu> -======================= Irfan Pyarali <irfan@cs.wustl.edu> - -This is a test which stress tests the connection management and checks -whether cached handlers are purged effectively when the process runs -out of descriptors. - -Connection Caching Feature --------------------------- - -Motivation: - -In TAO, open connections are stored in a connection cache and reused -for the subsequent requests between the same client and server. This -is done to avoid the cost of establishing new connections. Every -connection needs a OS handle which has to obtained from the OS. The OS -can allot only a fixed number of handles to a process. This can -become a issue in case of clients and servers with multiple handles -open who may not be allowed to establish a new connection as the -process has run out of handles. - -Aim: - -Clients and servers should be able to to continue making new -connections by tearing down "old" connections. - -Solution: Auto-purging - -The connections are stored in a cache. When ever the connect call -fails due to the process running out of handles, auto-purging will -kick in and some "old" connections will be removed from the connection -cache. The decision about which connections are "old" is made by the -caching strategies. The percentage of existing connections to be -purged is controlled by a purge percent. - -ACE classes: Caching_Utility, Cleanup_Stratgies, and -Caching_Strategies have been used. Simple examples are available in -$ACE_ROOT/tests/Cached_Accept_Conn_Test.cpp and -$ACE_ROOT/tests/Cached_Conn_Test.cpp. - -Note that the users can also purge handles used by the ORB explicitly -through TAO_Connector_Registry::purge_connections(). This can enable -them to open new handles for their application. - -Test: ----- - -This test creates the number of "server" ORBs specified by the -i -option and has a client ORB which invokes remote methods on these -ORBs. - -Two possible cases: - -1. When the process runs out of handles during "connect", the -Connector removes handlers from its connection cache depending upon -the kind of connection caching strategy followed. (by default: Least -Recently Used) set by the ORBConnectionCachingStrategy option. - -Its expected to see an exception when the handle limit is surpassed. -The cached handles are then used and the test continues displaying -how scalability was increased using caching of connection handles. - -2. When the "accept" or "open" calls made by the Acceptors in the same -process fail, request is made on all the Connectors of all the ORBs in -the system to perform purging depending on the ORBPurgePercentage -value provided. - -Note: This test does not use the Collocation optimisations since we -want to use up handles and test the connection caching as well as -purging. - -Options: -------- - -i: Number of iteration which equals the number of server orbs. - -d: Debugging ON - -a: Number of handles to be kept free. The Handle Gobbler will - eat up the max_handles specified for the OS but leave - behind the handles requested. This helps in testing more - effectively. Please remember to set this value judiciously. - -r: Number of remote call invocations to be made by the client - on the servant. - -?: help! - - -Options Using svc.conf: ------------------------ - -Through options to the Resource_Factory :- --ORBConnectionCachingStrategy <lru/lfu/fifo/null> --ORBPurgePercent <percentage> - -Execution: ---------- - -Simply run the ./run_test.pl ! - -else - -./Connection_Purging -ORBCollocation no - -with options specified per need. - - |